As of March 23 2021, there are 123 million confirmed COVID-19 cases and 2.7 million deaths reported to WHO for the 194 Member States.
There are significant differences in how COVID deaths are reported between countries (https://observatoryihr.org/news/chinarussia-brazil-and-the-underreporting-of-covid19-cases)
Variations in testing access, diagnostic capacity and how COVID deaths are defined
Deaths may be misattributed to COVID-19 and, converse
Excess mortality provides an estimate of the additional number of deaths for specified location and time period, compared to the number of deaths expected (often estimated using the same time period in the preceding year or averaged over several preceding years).
encompasses deaths from all causes
gives a summary measure of the ‘whole system’ impact.
Overall for the year 2020:
Global: 1.87 million COVID-19 and 3.75 million estimated excess deaths (UI 3.23-4.28)
AFRO: 45 thousand COVID-19 and 196 thousand estimated excess deaths (UI 114-278)
EMRO: 122 thousand COVID-19 and 320 thousand estimated excess deaths (UI 281-360)
EURO: 602 thousand COVID-19 and 1.1 million estimated excess deaths (UI 1.1-1.11)
PAHO: 896 thousand COVID-19 and 1.31 million estimated excess deaths (UI 1.29-1.32)
SEARO: 186 thousand COVID-19 and 638 thousand estimated excess deaths (UI 454-825)
WPRO: 20 thousand COVID-19 and 191 thousand estimated excess deaths (UI -14-399)
Countries (and data sources) that can be used to estimate excess mortality directly (Tier 1),
Approaches to indirectly estimate excess mortality using alternative data sources or models (Tier 2)
The most appropriate statistical models to estimate excess mortality in countries for which it cannot be determined empirically, directly or indirectly) (Tier 3). Considerations for the variable of interest, statistical model functional forms, best predictors/covariates. (Tier 3)
functional form eg \(\frac{d^o - d^e}{covid}\) vs \(\frac{d^o}{d^e + covid}\)
covariates eg historical NCD/COMM rates, age-patterns, population density, income, policies
Propagate uncertainty from background model